parents, schools and human capital differences across ......relatedliterature...

Parents, Schools and Human CapitalDifferences across Countries

Marta De Philippis† Federico Rossi‡

†Bank of Italy

‡Warwick University

Journal of the European Economic AssociationFinal Set of Slides: July 2020

Introduction

§ Human capital varies greatly across countries (years ofschooling, test scores..)

§ Recent growth literature Ñ important for cross-countrydifferences in income (Hanushek et al, 2000, 2009, 2012)

§ What explains differences in human capital achievement?

§ For test scores, policy debate mostly focused on schoolquality

§ At individual level, we know that home environment andparents are crucial for skill formation

§ Is parental influence important for cross-country gaps inperformance?

Quantifying Cross-Country Gaps in Parental Influence

§ Previous work Ñ include measures of parentalsocio-economic background when estimating “educationalproduction functions” (Woessman, 2016)

§ Several channels of parental influence not perfectlycaptured by those: parenting style, transmission ofpreferences and cultural traits..

§ Our proposal: study the school performance of secondgeneration immigrantsÑ compare children born and educated in the same

country/school, with parents from different nationalities

§ Performance gaps across parental nationalities as proxiesfor cross-country gaps in an unobservable parentalcomponent

What We Do

§ Look at two schooling outcomes:§ PISA score§ grade retention from US Census

§ Study the performance of second generation immigrants

§ Augment the "educational production function"specification with parental country of origin fixed effects,identified from second generation immigrants

§ Quantify the role of this parental component forcross-country gaps in PISA performance

§ Use data on parental migration history, education and timeuse to explore the nature of this parental component

Preview of Main Findings

§ Second generation immigrants from high PISA countries dobetter than their peers from low PISA countries

§ Patterns of selection on observables suggest that this resultis not due to differential selection into emigration

§ Unobserved parental characteristics account for§ 15% of the total cross-country variation in PISA scores (as

much as observable parental characteristics)§ More than 50% of the out-performance of East Asian

countries (compared to the average)

§ Evidence in supporto of a “cultural” explanation

§ Differences in parental time use across nationalities

Related Literature

§ Cross-Country differences in human capital: Bils andKlenow (2000), Hendricks (2002), Hanushek andWoessman (2012), Schoellman (2012, 2016), Manuelli andSeshadri (2014), Restuccia and Vandenbroucke (2014),Woessman (2016), Doepke and Zilibotti (2017)

§ School performance of immigrants: Levels et al (2008),Dustmann et al (2012), Jerrim (2014), Dronkers et al(2016)

§ Epidemiological approach and culture: Giuliano (2007),Alesina and Giuliano (2010), Fernandez and Fogli (2009),Fernandez (2011), Figlio et al (2019)

Outline

§ The Role of Parental Observable Characteristics

§ Second Generation Immigrants: Motivating Evidence§ PISA§ US Census

§ Selection into emigration

§ The Role of Parental Unobservable Characteristics

§ Mechanism

§ Conclusions

Data - PISA

§ Micro-data from PISA: cross-country evaluation ofstudents’ performances in math, reading and science

§ Repeated cross sections (2003, 2006, 2009, 2012, 2015)

§ Sample of 79 countries

§ In this presentation: results for math scores

Educational Production Function

§ Focus on “native” students

§ Generic educational production function

Ticst “ Parentsicst ` EduSystemcst ` β1Dicst ` αt ` εicst

§ Ticst : score in wave t of student i , in country c , school s

§ Parentsicst : effect of characteristics of parents and thehome environmentÑ Parentsicst “ ParentsObsicst ` ParentsUnobsicst

§ EduSystemcst : effect of resources and institutional featuresof the educational system

§ Dicst : student’s age, gender

Decomposing the Cross-Country Variation

§ Average score (across all waves) of students in country c :

Tc “ α`ParentsObsc`ParentsUnobsc`EduSystemc`β1Dc

§ Additively decompose the cross-country variance of Tc

§ Compute contributions of parental characteristics as

Covp {ParentsObsc ,Tcq

VarpTcq

Covp {ParentsUnobsc ,Tcq

VarpTcq

Educational System Controls


§ Various approaches to control for EduSystemcst :

1. Observable characteristics of schools and countries’educational environments (Woessmann, 2016): resources,accountability, monitoring, autonomy..Ñ available for 37 countries

2. Country ˆ Wave fixed effects

3. School ˆ Wave fixed effects

Parental Characteristics


§ The parental component is given by

Parentsicst “ ParentsObsicst ` ParentsUnobsicst

where ParentsObsicst “ ρ1Xicst

§ Xicst Ñ parents’ education, employment, occupationalstatus, number of books and language spoken at home

§ ParentsUnobsicst residual Ñ any systematic cross-countryvariation absorbed by country/school FE

Results - Baseline Decomposition

[1] [2] [3] [4] [5]

Covp {ParentsObsc ,Tc q

VarpTc q26.36 25.10 17.42 12.70 8.89

# Country 37 37 79 37 79School Controls Yes No No No NoHost Country ˆ Wave FE No Yes Yes No NoSchool ˆ Wave FE No No No Yes YesSample Edu Obs Edu Obs All Edu Obs All

The Role of Parental Observable Characteristics

§ Depending on sample and controls, parental observablecharacteristics account for 9%-25% of the cross-countryvariation in test scores

§ Larger sample: 9%-16%

§ Observables might fail to capture relevant channelsthrough which parents affect human capital accumulation

§ Might underestimate overall importance of parentalinfluence for cross-country gaps

Outline





§ Mechanism

§ Conclusions

Second Generation Immigrants

§ Laboratory to study cross-country differences in parentalinfluence

§ Two key premises1. Within host country/school, sec gen immigrants subject to

same educational system and country-level factors

2. Cross-country differences in parental practices and(unobservable) parental characteristics should be preservedacross countries of origin of immigrant parents

§ Test whether sec gen immigrants from high-PISA countriesto do better than peers from low-PISA countries

§ Caveats: emigrants’ selection, differential culturalassimilation..Ñ come back to those later

Data - PISA

§ Identify second generation immigrants based on students’and parents’ country of birth

§ Exclude students born in country different from the onewhere they take the test

§ We observe „ 50000 sec gen immigrants, across 39 hostcountries and from 59 countries of origin

§ Large heterogeneity in sample size by country of origin

§ “Core sample”: 31 countries of origin from with at least100 emigrant mothers and fathers

The Raw Data

ALBARG

AUS

AUT

BEL

BRA

CHN

HRV CZEDNKFRADEU

GRC

HKG

INDIRL

ITALBN

MAC

NLDNZLPOLPRT

RUSSVK

VNM

ESPSWE

CHE

TUR

GBR

USA

SCG

-1-.5

0.5

11.

5

Aver

age

Scor

e Se

c G

en Im

mig

rant

s on

Mot

her S

ide

-1 -.5 0 .5 1 1.5Average Score Natives

Specification

§ Estimate the following specification:

Tmicst “ θ0 ` θ1T

m ` θ12Xicst ` θcst ` εmicst

§ Tmicst Ñ score of student i , in school s, country c , wave t

whose mother is from country m§ Tm Ñ average score of native students in the mother

country of birth m§ Xicst Ñ socio-economic and demographic characteristics

(including dummy for native fathers) List

§ θcst Ñ country/school ˆ wave fixed effects

§ Sample = Second generation immigrants on mother’s side§ Standard errors clustered at the mother’s country of birth

level, inflated by estimated measurement error

Results - Math Score

Dependent Variable: PISA Math ScoreNo East

All All All All AsiaScore Country m 0.755˚˚˚ 0.628˚˚˚ 0.271˚˚ 0.225˚˚˚ 0.174˚˚

p0.208q p0.223q p0.119q p0.072q p0.082qN 49097 49097 49097 49097 31347# Country m 59 59 59 59 52R Squared 0.10 0.23 0.34 0.66 0.62Socio-Econ Controls No Yes Yes Yes YesHost Country ˆ Wave FE No No Yes Yes YesSchool ˆ Wave FE No No No Yes YesWave FE Yes Yes Yes Yes Yes

Robustness of Baseline Result

§ Reading and Science Show

§ Second generation immigrants on father’s side Show

§ Full sample (including natives) Show

§ Excluding source and host countries Show

§ More controls at the country-of-origin level Show

Data - US Census

§ Follow Oreopoulos and Page (2006): measure of graderepetition from children’s age and grade attended

§ No Grade Repeated = 1 if grade attended ě mode for thestudent’s state, age, quarter of birth and census year cell

§ 1970-1980 waves; focus on 8 to 15-year-old children

§ Useful information on parents’ immigration history

§ No school identifiers Ñ control for Commuting Zone fixedeffects

Results - US Census

Dependent Variable: 1= No Grade RepeatedNo East

All All All All AsiaScore Country m 0.094˚˚˚ 0.050˚˚˚ 0.032˚˚˚ 0.029˚˚˚ 0.023˚˚

p0.031q p0.014q p0.010q p0.010q p0.010qFemale 0.069˚˚˚ 0.069˚˚˚ 0.068˚˚˚ 0.068˚˚˚ 0.071˚˚˚

p0.003q p0.003q p0.003q p0.003q p0.003qMother Sec Edu 0.055˚˚˚ 0.047˚˚˚ 0.044˚˚˚ 0.042˚˚˚

p0.012q p0.012q p0.011q p0.012qMother Ter Edu 0.060˚˚˚ 0.053˚˚˚ 0.050˚˚˚ 0.045˚˚˚

p0.011q p0.009q p0.010q p0.010qFather Sec Edu 0.045˚˚˚ 0.039˚˚˚ 0.040˚˚˚ 0.045˚˚˚

p0.014q p0.010q p0.010q p0.009qFather Ter Edu 0.063˚˚˚ 0.062˚˚˚ 0.063˚˚˚ 0.068˚˚˚

p0.015q p0.012q p0.011q p0.011qLog Family Income 0.043˚˚˚ 0.036˚˚˚ 0.034˚˚˚ 0.035˚˚˚

p0.010q p0.008q p0.008q p0.008qN 53553 53553 53553 53553 49634# Country m 64 64 64 64 57Years Since Migr Mother no no no yes yesComm. Zone FE no no yes yes yes

Other controls: child’s and parents’ age dummies, family size, year FE, quarter of birth FE,father’s immigrant status.

Outline





§ Mechanism

§ Conclusions

Selection

§ We compare second generation immigrants across parentalnationalities

§ A common degree of selection across countries of originwould not be a problem

§ We worry about the possibility of differential selectionacross high- and low-PISA countries

§ In particular, more positive selection from high-PISAcountries could rationalise our results

Selection on Observables

§ Build proxies of selection based on observablecharacteristics

§ Parental education:

Selectionmi ,c “YrsEdumi ,c ´ YrsEdu

mm

SDpYrsEdumi ,mq

Selection into Emigration - Evidence for Mothers

ALB

AZE

ARG AUSAUT BEL

BRA

BGRCANCHL CHN

TWN

COL

HRV CZE

DNK

ESTFINFRA

GEODEU

GRC

HKGHUN

ISLIND

IRL

ITAJPN

KAZ

JORKOR

XKXLBN

LIE

MAC

MYS

NLD

NZL

NOR

PANPOL

PRTROURUS

SGPSVK

VNMSVN

ESPSWE

CHE

THA

TUR

MKD

GBR

USAURY

SCG

-2-1

01

2Av

erag

e St

anda

rdize

d Ed

ucat

ion

of E

mig

rant

Mot

hers

-1 -.5 0 .5 1 1.5Average Score Natives

Selection into Emigration - Regression Evidence

Dependent Variable:Standardized Years of Education

[1] [2] [3] [4]Mothers Fathers

Score Country m 0.005 -0.083(0.222) (0.197)

Score Country f 0.019 -0.135(0.210) (0.164)

N 49097 49097 48834 48834R Squared 0.07 0.61 0.07 0.59Host Country ˆ Wave FE Yes No Yes NoSchool ˆ Wave FE No Yes No Yes

Selection - Taking Stock

§ No evidence of positive differential selection

§ Consistent with findings in development accountingliterature (Hendricks and Schoellman, 2018)

§ If anything, migrants from poorer (and low PISA)countries more positively selected

§ Similar results from the US Census sample Show

§ Results robust to controls for linguistic and culturaldistance Show

Outline





§ Mechanism

§ Conclusions

Educational Production Function

§ Re-consider the educational production function, including bothnatives and second generation immigrants

Tmficst “ Parentsmf

icst`αcst`µ1Dicst`θ

mNatMothmicst`ζfNatFathficst`ε

mficst

§ Tmficst : score in wave t of student i , in country c , school s, with

parents born in countries m and f

§ Parental component:

Parentsmficst “ ParentsObsicst ` ParentsUnobsicst

“ ξ1Xicst ` γm ` δf ` umf

icst

where γm and δf are country-specific average unobservablecontributions of mothers from m and fathers from f

Educational Production Function - Estimation

§ γm and δf identified by parental country-of-origin fixedeffects, out of second generation immigrants

§ Estimate

{ParentsObsc “ pξ1Xc , {ParentsUnobsc “ pγc ` pδc

§ Compute contributions for cross-country variation as

Covp {ParentsObsc ,Tcq

VarpTcq,

Covp {ParentsUnobsc ,Tcq

VarpTcq

for countries in the Core Sample (at least 100 emigrantmothers and fathers)

Decomposition - Results

[1] [2]

Covp {ParentsObsc ,Tc q

VarpTc q21.21 10.64

Covp {ParentsUnobsc ,Tc q

VarpTc q15.86 11.57

# Country 31 31Host Country ˆ Wave FE Yes NoSchool ˆ Wave FE No YesSample Sec Gen Sec Gen

Decomposition - East Asia Countries vs Mean

{ParentsObsc Gap {ParentsUnobsc GapCountry PISA Gap School FE Country FE School FE Country FE

Hong Kong 0.72 0.00 0.00 0.19 0.43(0.00) (0.01) (0.15) (0.15)

China 0.54 -0.04 -0.09 0.25 0.62(0.01) (0.01) (0.04) (0.05)

Macao 0.37 -0.04 -0.10 0.19 0.47(0.00) (0.01) (0.08) (0.10)

Vietnam 0.16 -0.14 -0.31 0.29 0.49(0.01) (0.01) (0.05) (0.06)

Outline





§ Mechanism

§ Conclusions

Mechanism

We proceed in three steps:

1. Heterogeneity analysis

2. Controls for cultural proxies

3. Parental time use

Heterogeneity: Parental Education

§ Possible interpretation Ñ Intergenerational transmission ofeducational quality

§ Would imply an even more central role for the schoolsystem

Ñ would imply a stronger country-of-origin effect amongparents with higher education in their home country

Heterogeneity: Mother’s Education

0.0

5.1

.15

.2Co

effic

ient

of A

vg S

core

in M

othe

r's C

ount

ry o

f Orig

in

NoSchooling

Primary Secondary Some College College>= 4 yrs

Mother's Education

Heterogeneity: Years Since Migration

§ Possible interpretation Ñ Country-specific Cultural Traits

§ Country-of-origin effect should attenuate as parentsintegrate in the US

Ñ weaker country-of-origin effect among parents who havespent more years in the US

Heterogeneity: Years Since Migration

-.05

0.0

5.1

Coef

ficie

nt o

f Avg

Sco

re in

Mot

her's

Cou

ntry

of O

rigin

<15 15-19 20-24 25-29 30+Mother's Years Since Migration

Proxies for Culture

Five proxies from the World Value Survey:

§ Long-term orientation (Dohmen et al, 2015; Galor andÖmer, 2016; Figlio et al., 2019)

§ Beliefs of importance of hard work (Weber, 1930)

§ Locus of control (Coleman and DeLeire, 2003; Lekfuangfuet al., 2018)

§ Trust (Guiso et al., 2006; Algan and Cahuc; 2014;Coleman, 1988; Rafael La Porta and Vishny, 1997;Bjornskov, 2009)

§ Prevalence of secular and rational values (Inglehart andWelzel, 2005; Ek, 2018)

Proxies for Culture

Dependent Variable: Math Score(1) (2) (3)

Score Country m 0.223˚˚˚ 0.038 0.111p0.073q p0.082q p0.081q

Long Term Orientation 0.464 ˚ ˚ 0.264˚p0.215q p0.137q

Hard Work ´0.091p0.124q

Trust 0.193p0.148q

Locus of Control 0.372 ˚ ˚ 0.305˚˚˚p0.162q p0.106q

Secular-Rational Values ´0.078p0.060q

N 48398 48398 48398# Country m 52 53 52Socio-Econ Controls yes yes yesHost Country FE yes yes yesSchool FE yes yes yesYear FE yes yes yes

Time Use

What do parents from high PISA countries do differently?

§ Look at immigrant parents in the ATUS Survey, 2002-2015

§ Total time spent in childcare, split between educational,recreational and basic activities (following Aguiar andHurst, 2007)

§ No school identifiers Ñ control for State FE

Time Use of Parents

Total Educational Recreational BasicScore Country p 11.966˚˚˚ 3.118˚˚ 5.866˚˚ 2.982

p4.258q p1.333q p2.271q p2.046qMother 66.230˚˚˚ 8.480˚˚˚ 1.090 56.660˚˚˚

p3.854q p0.870q p3.192q p2.434qParent Sec Edu ´1.702 4.631˚˚˚ ´2.945 ´3.389

p6.002q p0.764q p3.170q p2.638qParent Ter Edu 4.220 4.100˚˚˚ ´1.999 2.119

p3.586q p1.297q p2.288q p1.850qSpouse Sec Edu 3.242 ´1.869˚˚ 6.188˚˚ ´1.077

p2.912q p0.813q p2.832q p1.225qSpouse Ter Edu 12.816˚˚˚ 2.022 6.736˚˚ 4.059

p3.388q p1.517q p2.655q p3.016qN 5812 5812 5812 5812# Country p 64 64 64 64Mean Dep. Var. 89.33 10.54 22.03 56.75State FE yes yes yes yesYear FE yes yes yes yes

Other controls: years since migration, age, number of children, children’s gender andage, father immigrant status, log family income, dummies for retired, disabled, full timestudents. Standard errors clustered by parent’s country of origin.

Outline





§ Mechanism

§ Conclusions

Conclusions

§ Empirical strategy to infer the importance of unobservableparental characteristics for cross-country differences inhuman capital

§ Importance of parents goes well beyond the impact of theirsocio-economic characteristics

§ Variation in parental influence related to cultural factors

Ñ Macro models of human capital accumulation should beconsistent with cross-country differences in the role ofparents

Ñ Policy: importing school practices of high PISA countriesmight not be enough to improve test performance

List of Socio-Economic Controls

§ Student: age in months, gender

§ Parents: parental education dummies, employment statusdummies, ISEI index of occupational status, number ofbooks at home dummies, different language spoken athome dummy, father’s immigrant status dummy

Back

Results - Reading Score

Dependent Variable: PISA Reading ScoreAll All All All No Asia

Score Read Country m 0.600˚˚ 0.409˚ 0.095 0.143˚˚˚ 0.112˚˚p0.248q p0.212q p0.091q p0.047q p0.048q

Female 0.296˚˚˚ 0.264˚˚˚ 0.255˚˚˚ 0.208˚˚˚ 0.229˚˚˚p0.034q p0.028q p0.023q p0.028q p0.029q

Father Sec Edu 0.039 0.061˚˚ 0.055 0.113˚˚˚p0.056q p0.031q p0.038q p0.033q

Father Ter Edu ´0.049 0.085˚˚ 0.049 0.093˚˚p0.077q p0.038q p0.041q p0.045q

Mother Sec Edu 0.072 0.090˚˚ ´0.023 ´0.003p0.072q p0.042q p0.026q p0.047q

Mother Ter Edu ´0.044 0.110˚˚˚ ´0.017 ´0.005p0.095q p0.039q p0.035q p0.057q

N 49097 49097 49097 49097 31347# Country m 59 59 59 59 52Host Country ˆ Wave FE No No Yes No NoSchool ˆ Wave FE No No No Yes Yes

Other controls: father immigrant status, book at home, parents’ working status,language spoken at home.

Results - Science Score

Dependent Variable: PISA Science ScoreAll All All All No Asia

Score Science Country m 0.711˚˚˚ 0.507˚˚ 0.227˚˚ 0.245˚˚˚ 0.209˚˚˚p0.240q p0.228q p0.115q p0.068q p0.076q

Female ´0.038 ´0.070˚˚ ´0.082˚˚˚ ´0.125˚˚˚ ´0.103˚˚˚p0.037q p0.030q p0.026q p0.026q p0.023q

Father Sec Edu 0.046 0.079˚˚ 0.066˚ 0.122˚˚˚p0.064q p0.033q p0.034q p0.040q

Father Ter Edu ´0.021 0.123˚˚˚ 0.084˚˚ 0.126˚˚˚p0.074q p0.028q p0.040q p0.049q

Mother Sec Edu 0.005 0.063 ´0.023 0.013p0.068q p0.038q p0.030q p0.050q

Mother Ter Edu ´0.111 0.086 ˚ ˚ ´0.024 ´0.002p0.091q p0.037q p0.033q p0.057q


Other controls: father immigrant status, book at home, parents’ working status,language spoken at home.

Back

Results - Fathers

Dep Variable: PISA Science Score(1) (2) (3) (4) (5)

Score Country f 0.792˚˚˚ 0.653˚˚˚ 0.305˚˚ 0.202˚˚ 0.148p0.194q p0.215q p0.132q p0.085q p0.096q

Female ´0.113˚˚˚ ´0.142˚˚˚ ´0.156˚˚˚ ´0.199˚˚˚ ´0.183˚˚˚p0.035q p0.034q p0.030q p0.026q p0.029q

Father Sec Edu ´0.063˚˚ ´0.024 ´0.005 ´0.004p0.030q p0.028q p0.017q p0.035q

Father Ter Edu ´0.134 ˚ ˚ 0.003 ´0.003 ´0.010p0.055q p0.043q p0.037q p0.054q

Mother Sec Edu 0.078 0.088˚˚ ´0.009 0.041p0.060q p0.039q p0.041q p0.077q

Mother Ter Edu ´0.025 0.106˚˚˚ 0.009 0.050p0.072q p0.038q p0.040q p0.077q


Back

Results - Full Sample

Dep Variable: PISA Math Score(1) (2) (3) (4) (5)

Score Country m 0.414˚˚ 0.371˚˚ 0.229˚ 0.150˚˚ 0.105˚p0.177q p0.172q p0.123q p0.067q p0.059q

Score Country f 0.459˚˚ 0.403˚˚ 0.248˚˚ 0.160˚˚ 0.102p0.180q p0.177q p0.118q p0.065q p0.063q

Score Country m * Native Mother 0.174 0.067 ´0.031 ´0.032 0.055p0.149q p0.163q p0.098q p0.059q p0.051q

Score Country f * Native Father ´0.026 ´0.088 ´0.176 ´0.114˚˚ ´0.035p0.153q p0.167q p0.112q p0.057q p0.060q

Female ´0.110˚˚˚ ´0.113˚˚˚ ´0.112˚˚˚ ´0.143˚˚˚ ´0.145˚˚˚p0.012q p0.012q p0.012q p0.013q p0.015q

Native Mother ´0.117˚ ´0.179˚˚ ´0.052 ´0.013 0.035p0.064q p0.086q p0.060q p0.040q p0.032q

Native Father ´0.009 ´0.142˚ ´0.028 0.004 0.009p0.069q p0.076q p0.066q p0.029q p0.029q

N 1445071 1445071 1445071 1445071 1326079# Country m 59 59 59 59 52# Country f 58 58 58 58 52Host Country ˆ Wave FE No No Yes No NoSchool ˆ Wave FE No No No Yes Yes

Back

Omitting Countries of Origin

0.1

.2.3

.4.5

Coef

ficie

nt

ALBARGAUSAUTAZEBELBG

RBRACANCHECHLCHNCO

LCZEDEUDNKESPESTFINFRAG

BRG

EOG

RCHKGHRVHUNINDIRLISLITAJO

RJPNKAZKO

RLBNLIEM

ACM

KDM

YSNLDNO

RNZLPANPO

LPRTRO

URUSSCGSG

PSVKSVNSW

ETHATURTW

NURYUSAVNMXKX

Excluded Country of Origin

Back

Omitting Host Countries

0.1

.2.3

.4.5

Coef

ficie

nt

ARGAUSAUTBELCHECRICZEDEUDNKDO

MESTFING

BRG

EOG

RCHKGHRVIDNIRLISRKAZKG

ZKO

RLIELUXLVAM

ACM

DAM

EXM

KDM

USNLDNO

RNZLPRTQ

ATSCGSVKSVNTURURY

Excluded Host Country

Back

Country-Level Controls

Dependent Variable: PISA Math Score(1) (2) (3) (4) (5) (6)

Score Country m 0.291˚˚˚ 0.228˚˚˚ 0.254˚˚˚ 0.230˚˚˚ 0.332˚˚˚ 0.351˚˚˚

p0.077q p0.085q p0.073q p0.082q p0.067q p0.070qExp.per Stud m ´0.007˚˚ 0.002

p0.003q p0.009qSome Shortage m 0.406˚˚ 0.254

p0.189q p0.264qLarge Shortage m ´0.241 ´0.431

p0.224q p0.464qPupil/Teacher m 0.003 0.014˚˚˚

p0.007q p0.005qAvg Years Edu m 0.001 0.017

p0.012q p0.013qLog GDP m ´0.114˚˚˚ ´0.144˚

p0.034q p0.083qN 48834 48834 48834 48834 48834 48834# Country m 49 49 49 49 49 49Socio-Econ Contr Yes Yes Yes Yes Yes YesSchool-Wave FE Yes Yes Yes Yes Yes Yes

Standard errors clustered at the mother’s country of origin level.

Back

Selection into Emigration - US Census

§ Measure average and standard deviation of 35-45 years oldadults’ education in country of origin from Barro and Lee(2001)

§ If anything, negative differential selection Show

§ Restrict attention to parents entirely educated in homecountry:

§ Similar pattern of differential selection Show

§ No differential gradient with respect to time betweeneducation completion and migration Show

Back

Selection into Emigration - US Census

ALB

DZA

ARG

AUS

AUTBEL

BRA

BGR

CAN

CHL

CHN

COLCRI

DNKDOM ESTFINFRA

DEUGRC

HKGHUNISL

IND

IDN

IRLISR ITA

JPN

JOR

KORLVALTU

LUX

MACMYS

MLT

MUS

MEX NLDNZL

NORPANPER

POLPRT

QATROU

SGPVNM

ESPSWE

CHE

THA

TTO

TUN

TUR

GBRURY

VEN

02

46

8Av

erag

e St

anda

rdize

d Ed

ucat

ion

of E

mig

rant

Mot

hers

-1.5 -1 -.5 0 .5 1 1.5Average Score Natives

Back

Selection into Emigration - Parents Educated in HomeCountry

ALB

DZA

ARG

AUS

AUTBEL

BRA

BGR CAN

CHL

CHNCOL

CRI DNKDOM ESTFINFRA

DEUGRC

HKGHUN

ISL

IND

IDN

IRLISR ITA

JPN

JOR

KORLVA

LTU LUX

MACMYS

MLT

MUS

MEX NLDNZL

NORPAN

PER

POLPRTQAT ROU

SGPVNM

ESPSWE

CHE

THA

TTO

TUNTUR

GBRURY

VEN

02

46

8Av

erag

e St

anda

rdize

d Ed

ucat

ion

of E

mig

rant

Mot

hers

-1.5 -1 -.5 0 .5 1 1.5Average Score Natives

Back

Selection into Emigration - Parents Educated in SourceCountry

Dependent Variable:Standardized Years of Edu

Mothers FathersScore Country m ´0.668˚

p0.382qScore Country m ˆ Years betw Edu and Migration Mother 0.014

p0.015qYears betw Edu and Migration Mother ´0.060˚˚˚

p0.007qScore Country f ´0.248

p0.381qScore Country f ˆ Years betw Edu and Migration Father ´0.462

p0.326qYears betw Edu and Migration Father ´0.055˚˚˚

p0.005qN 30118 26658R Squared 0.21 0.25Comm Zone FE yes yesYear FE yes yes

Back

Linguistic and Cultural Distance

§ Additional concern: differential quality of thecountry-student “match”

§ Linguistic Distance: constructed following the AutomatedSimilarity Judgment Program (Wichmann and Brown,2016)

§ Cultural Distance from Spolaore and Wacziarg (2015):based on the WVS

Linguistic and Cultural Distance

Dependent Variable: PISA Math Score(1) (2) (3) (4)

Score Country m 0.247˚˚˚ 0.277˚˚˚ 0.229˚˚ 0.287˚˚˚p0.075q p0.071q p0.096q p0.103q

Mother Linguistic Distance 0.018˚p0.009q

Father Linguistic Distance 0.017˚p0.009q

Mother Cultural Distance 0.035p0.024q

Father Cultural Distance 0.0095p0.026q

N 46896 46896 23513 23513# Country m 49 49 49 49Host Country ˆ Wave FE yes yes yes yesSchool ˆ Wave FE no yes yes yesSocio-Econ Controls yes yes yes yes

Back

parents, schools and human capital differences across ......relatedliterature...

Documents